Nikola Mrkšić, PolyAI: Enterprise AI sales, pricing, & millions in ROI

0:01What kind of software to the buy from what kind of people, right? And you have to match that and that will reduce the friction. Now, of course, maybe you are the god of sales and you've come up with a genius thing that's immediately obvious to everyone. If so, proceed. If not, you're probably going to have to learn about how software tends to be priced in what you're doing and you're going to have to mirror a lot of it, if not all of it, in the first instance. Welcome to the SaaS Revolution Show, a podcast by SaaS.com. Here we interview SaaS founders from around the world who've been there and done that.

0:34as they share the ins and outs of how they built their businesses, their operations, their path for securing investment, and more. Our mission with the podcast is to help you, the founder, learn how to scale your SaaS, maintain your well-being, and navigate the complexities of this ever-changing industry. I'm your host, Alex Deema, and together we'll explore the good, the bad, and the ugly in the journey to SaaS success. Welcome to the SaaS Revolution Show. I am your host, Alex Deema, CEO, founder of SaaSDoc, also general partner at Back Future Ventures. Delighted to be joined today ahead of his appearance and speaking at Sastoc Europe, which is on the 14th and 15th of October in Dublin, by the co-founder and CEO of Poly AI, Nicola Merksic.

1:18How are you doing, Nicola? I'm great. I'm great. Thank you for having me. Yeah, good to have you on the podcast. Obviously now having, I guess, a virtual conversation, but soon we'll get to meet in Dublin and very excited for you to come over and all the great, I guess, kind of like AI, you know, first and AI native, you know, founders and companies that will be there and, you know, getting together in person. Ahead of that, we have this opportunity to learn more about you and your story and building Poly AI, which is, you know, I think, It's certainly one of the most exciting AI companies in Europe and coming out of the UK.

2:04Nicola, I'd love to know, because I think you've got quite an interesting background coming from Belgrade. I understand that you, I think, wanted to be a writer, perhaps science fiction, maths, and going into Cambridge and how this spun out. So I'd love you to kind of share that story a little bit. Oh, for sure. Look, yeah, I think I said that in one podcast and that's become like the main team. No, no, look, I mean, I had a very classical kind of like Eastern European upbringing that, you know, maths all the way, STEM and nothing else exists, competing in maths, programming, all of those things.

2:41I think in that episode you know like I think the other fun part is you know there's classical Eastern Europe and then there's like former Yugoslavia Serbia in the 90s which is you know civil war land country falling apart a lot of instability so I think at one point I wrote five kind of like Star Trek inspired tribute SF sci-fi stories right and I had this idea of just going around like cafes and trying to like sell them to people see if someone reads them I don't really know what the end game was to be very honest with you but I came up with this like packaging where basically you could like buy three or five at different prices and I remember distinctly there was a day where I made like I think two or three hundred euros now the average Serbian salary that year the monthly salary was like two or three hundred euros so that felt like I was walking around with this bag full of money and I was like there's a way to hack it and I think you know a lot of startup growth and all of that almost feels the same in that there's a lot of theory and you know what the averages are what slightly better is and then sometimes you just have these breakthroughs and sometimes it's way harder than the textbook says and sometimes it you know you just find an opening and then you got to use it because god knows it's not easy building those things as well so uh you know i ended up at cambridge doing uh computer science and maths um towards the end of kind of like my master's and i met a guy who was starting a company called vocal iq uh he dropped out of a cambridge lectureship to do that and he was starting with another cambridge professor so So a very classical Cambridge story where it's kind of like, you know, a spin out of people who were among the first to do deep learning for voice assistants back when no one cared about it.

4:17Right. They had been working on it before Siri was a thing. Right. So that company ended up acquired by Apple a year and a half later. And I ended up doing a PhD with Steve Young, one of the two co-founders and a legend in the field. And then deep learning really took off. right so whatever you were doing i just started working 10 times better every year and it hasn't really stopped ever since but what got me interested in building a company that would kind of like do b2b um agents for customer service which you know i wouldn't have told you i was going to do 10 years before right um was the fact that you know when apple acquired us people would ask where do you work apple what do you work on siri serious shit and you go okay okay like fair some things didn't work some things did but there was more than just oh i said this and it wasn't understood i think the mismatch between our sci-fi style expectation of what this assistant would be and how we would use it and the actual reality was so wide that you could be nothing but disappointed and should be told it was less to do with siri herself and more to do with the expectations.

5:30So to me, kind of like the methodical approach was let's narrow the problem down to something a bit more manageable where we can both hopefully do it way better relative to expectations and measure whether we're doing well, improve every year. So to build a company, you need like to delight customers, get more usage, get paid for it, get data, improve the product, delight them even more. So the cycle there that was very obvious was customer service because believe it or not, today, as well as five years ago, about 2 % of the working population in the UK and the US works in contact centers. So despite all the automation, all the deflection, all the chatbots, it hasn't really shifted all that much.

6:13And that's ignoring the probably equal amount of labor that offshore support enterprises in Western societies around providing customer service. So So, you know, it takes a village to provide a customer service to your customers. And most companies today are doing it quite badly. And AI has a lot to give there. But, you know, it has yet to give a lot, right, in terms of just transforming that whole estate. Is that 2 % you think going to become 1 % and 0.5 % because of AI? Is that how you would see it potentially? I think so. I think it will lead to definitely a 5x reduction over some span of time, whether it's three to five years remains to be seen.

6:59It's not that important. I think what I'm really excited by and that we're witnessing in a lot of our deployments is that it creates a new generation of knowledge workers. It gives like social mobility to the best, brightest, hardest working people in the contact center. that's historically been quite hard for companies to do to identify people there and you know promote them up and extract talent because they get you know wherever you get a lot of people going through your company there's a lot of talent going through whether you're identifying it and moving it to the place where you need it most probably not but um if you have them you should identify them and i think that with um the context center transforming into this place where you have humans and AI, not working really side by side, but working on different things, the best paid job and the most exciting job becomes that of the future air traffic control for AI.

7:51Because if you have AI doing 80 % of the work, any mistake, well, like the upside is huge, not to mention the cost savings and the ability to pick up the phone or answer a chat request anytime of day or night in whatever language, you know, the spiel. it's really that if something goes wrong it goes wrong at a scale that humans could never pull off so it really will start to resemble air traffic control where it's going to pay to have levels and levels of supervision and just continuous improvement both for the sake of improvement but also for the sake of having assurances that you don't start you know issuing product returns to people who called you about you know an upsell or something like that you started the business in 2017 is is that right yeah and i think perhaps like back then we call it like machine learning deep learning is like you know conversational you know intelligence perhaps still would you say though that over the last couple of years i mean obviously with this fast pace of you know the new ai like developments like has there really been like an inflection point like a a wave that has you know in a positive way like impacted you as a business you know are you able to correlate that 100 look i think the first three four years we were building prototypes and really it was we were what you would look at as a classical research lab right When the ratio of PhDs and postdocs to employees is over 50%, it's a fake company when it comes to really being a company, right?

9:28It's something else. Really, it's around 21 only that we had the first pilots and things running. And then it really started growing. And it was growing fast even pre the ChowGPT moment. And then it exploded when Sam Maltman and Ilya did their magic. Because not only did the capabilities start evolving. They both jumped in like what an LLM could do is very different from what you could do with the previous machine learning approaches. But more importantly, I don't know if it's more important, it's equally important. The interest of enterprises and their belief that this can be done with technology has changed profoundly.

10:05And it shifted from early adopters pressing and finding for wiggle room inside their organizations to buy this kind of software and implement it. into boardrooms screaming at people about not implementing AI already as if they themselves knew how. But that's good because it just gives you a way to not do pilots in the fringe of a company and then look for other ways and other backers to kind of sponsor a wider rollout, which we have at this point gotten really good at doing. I mean, there's still a level of it, right? But now when the CIO is bang on the table and demanding that AI be implemented and you are a credible vendor with case studies, things are never easy for a growth stage startup, but man, it's a lot easier than it was.

10:49What about maybe some of the challenges around the pace of development, right? Because I guess what is true today won't necessarily be true in a couple of months' time. It's just going so fast. So you may have to rewrite parts of the product. And I don't know how you're finding that internally within the organization and for yourself. It's the challenge of building on something today and then in two months' time it could be something else and you've probably already experienced this. Oh, absolutely. Look, on the one hand, it's what we've been preparing for our whole lives because I think we are unique among our broader set of competitors and there were actually researchers who worked on the model front and then we went into the enterprise trenches, trenches fighting for adoption and clients and wide rollouts.

11:35So I think we have a better intuition than any of our competitors around how every change will impact the stack. Now, you know, I was at the OpenAI launch party at NIPS long ago and spoke with Ilya about like, you know, how much structure you should have in neural nets. And he, of course, insisted you should have none. And it's all about data. And, you know, even though I was a deep learning zealot, I thought he was like really out there in terms of the ambition. He was right about everything, right? So to say that we know what's going to come next would be foolhardy. We do not, right? I think we can talk about fine-grained assumptions and what we think will happen.

12:13But at this point, we might as well be talking about tonight's football game and how I think it's going to go, right? So I think you just have to be agile. The challenge, I think, in a B2B company, and frankly, even for OpenAI, you saw it with the GBT5 launch, is you try to tie your user base to your new model and to the next thing and the next thing. But it will always break some assumptions that are important to different parts of your customer base. And then, you know, it's a one by one transition for all of them. Sometimes it's a flip of a switch. Sometimes it is not. Sometimes you need their buy-in.

12:47Sometimes you don't. Sometimes it's really hard to do and you have to reimplement the whole thing. So the bigger the business, the slower you get. And that's where, you know, the heart of disruption always lies, where you have to be as fast as you can be. And it is inversely correlated with how big you are as a company. So, you know, we're sizable, but still quite small. So we're managing. but um you know it gets harder and harder the more clients we have what about uh for yourself personally is i guess kind of coming out of you know uh the research labs being this technical founder creating very technical product then going to have to sell this to to enterprise um you know what were you some of your learnings you know from that um that you could share yeah listen i think in europe we have this learn helplessness tech founder la la la like you get through to far worse in Europe, being a tech founder and everyone assumes that you're this moron who can't sell, talk, or do anything else, and you're expected to stay in your hole.

13:41In America they celebrate tech founders, and they like them more as they should, because those are full-star people who, you know, if they can't sell, they can't sell. They shouldn't be like CEO or whatever, right? I never really thought of myself as, I mean, I obviously am a deeply technical founder, right? But equally, you know, if you're investing in any one of these, like, Eastern Europeans that went to these specialist schools and moved on, know that they had to do that to get to where they were, right? That was the ticket. So much like in America, you get a lot of people who play like college football and then never look back.

14:13You know, I think like competing in maths for a lot of the former Soviet bloc is literally the same thing as Americans playing sports to get a scholarship. So I was never really, you know, had that hell bent on being technical, obviously, I know a lot about it and enjoy it. But yeah, to learn how to do it I think like you know someone who studied something non-technical has to learn how to build and scale an enterprise software company all the same you know I don't think there's a college degree they'll teach you that key insights I don't know I think it's just like founder sales all the way you have to lean in because you are initially like a one-man then maybe a two or three not very large group of people that is at the heart of looking for that product market fit, right?

14:59As you're selling, you're tweaking the product and the product offering. Half of it is probably not even built at the time that you're selling the first thing. So you have to have a very quick iteration loop. And the best way to have that is to have a single person or a few of them tightly integrated, owning the whole thing. And, you know, I think a cardinal sin, especially in Europe, is this whole like, oh, Luke, technical founder, now let's hire them. Killer head of sales. The truth is the killer head of sales who can really sell any software easily. A, there aren't many of them. B, they're working at very large companies, making seven-figure, like, you know, paychecks every year.

15:37And why would they, you know, go through, basically being co-founder with that person early on. So there's no silver bullet you're going to have to do it on your own. You should do it on your own. That's actually where the fun is. In terms of, so you're selling to the enterprise. in terms of coming up with the pricing for poly AI, have you done this based on outcome-based pricing, looking at these are the savings that we're going to make you as a customer by replacing human labor and perhaps software in that? How have you come to the price points which you're at at the moment? That's all very nice and theoretical.

16:17If I knew everything about your business, I could price it fairly. if you were willing to share it all with me, if you even knew it yourself. And if you were the decision maker in charge of making that decision and betting on the right outcome-based thing, there are companies that think that way and they love it priced that way. And we love pricing it that way because we make way more money, right? And the incentives are aligned and it all makes sense. A lot of others look at it as, you know, an analogous thing to the picks and shovels they can buy from the hyperscalers and others and have their IT teams build.

16:50and to those guys you will like the customer is always right so you will sell in the way that the customer wants to buy like about half of our contracts if maybe even 60 percent are consumption based the rest are outcome based and you know that outcome is so heterogeneous that i think there's a bit too much excitement about it now as if it's just a silver bullet it is not it is really hard to price outcome-based things the right way when you can do it it's magic but equally i think you would be mistaken to be an early stage company who thinks that you know just saying that it's going to be outcome-based you don't pay when it doesn't work it's like they're paying you and spending time with your worthless little company that has proven nothing yet right so or proven very little so you know that your enterprise like respecting an enterprises to understand that them spending time with you and going through all this is an investment, a serious one that they've prioritized over many other things that they probably need to do, right?

17:52So they're already paying for working with you and they've already put you ahead of many other things that they're no doubt being asked to do. Not just your competitors for your single thing, but many other things, right? So understanding that is, I think, where a lot of early stage guys really just are narcissistic. And it's like, but me, me, me, if I give it to you for free and it doesn't work well you know work with me because i could be outside cleaning my shed instead right and maybe that will give me higher roi maybe changing my erp system or my crm or god knows what else will be higher roi so i think there we just really have to think about what is your product and how do people like the people you're selling to what kind of software to they buy from what kind of people right and you have to match that and that will reduce the friction of course maybe you are the god of sales and you've come up with a genius thing that's immediately obvious to everyone if so proceed if not you're probably going to have to learn about how software tends to be priced in what you're doing and you're going to have to mirror a lot of it if not a little bit in the first instance your um your talk at sas.europ in a couple of weeks time is lessons in scaling enterprise ai how poly ai made voice agents deliver real roi um can you give a couple of insights because we obviously don't want to you know give the whole talk away here uh nor do we have time but um on you know some of like how you've been able to use voice agents to deliver uh roi and also what people can expect from your talk yeah yeah i mean look i think one of the main insights has been that people that are going to use ai to make their company the best version of itself are not starting with that whole oh i'm going to cut some costs because in the hierarchy of needs when selling cost is like a distant fourth right a dollar earned is a lot better than a dollar saved compliance and risk also rate i rank more highly because you get fired if you mess something up and then cost savings are nice to have so our explosive product market fit started in hospitality first we sold to restaurant groups and casinos one in two groups in vegas runs poly i right now why because if they don't pick up the phone and it's someone calling to make a booking either they will not make the booking or they'll make it online through an online travel agency and they'll lose 20 to 30 percent so we're just helping them make more money and that just like all the money right so there we've had outcomes like you know per casino site we increased like annual revenue by five to ten million for large restaurant groups again like tens of millions of revenue uplifts and that's just like it stops being negotiated like if it works it works it's worth trying the other part maybe that's interesting for that audience is just it happens a bit more easily in America, both because there's a lot more money sloshing around and because culturally Americans tend to be a bit more, you know, free thinking.

20:43They allow mid to kind of mid-high level executives to make calculated bets and just see whether it works or not. And if it does, great. And if not, so what? Whereas I think in Europe, there's been more of the consensus, blame game around why did you spend that money? Did we all sign up to it? And that just creates friction. Now, when you look at our largest deployments and, you know, beyond hospitality, it's really about improving customer experience in one shape or another. It's large enterprises whose customer service collapsed because they just can't have enough agents. They can't train them well enough, right?

21:15So their CSAT dropped, their waiting times got elongated. And they were like, what are we going to do? Let's throw AI at it, right? Especially today when AI is a lot more, it's more culturally accepted that AI will work and will continue to improve, right? So it's easier for them to make this bad. that's one piece the second piece is like just improving nps scores right once you show that for a bank for instance we improved nps score by 14 and that moves them a whole category of like what kind of bank you are to your customers and that's just so valuable that all the bait falls off right and i remember one of my early guys had this magic sentence he would walk in and just say we're going to cut 20 of your opex and i think the guy expected everyone to just go oh my god wow oh yeah, where do I sign?

21:57And that's where I just learned that that whole cost cutting and the way that I think the media and everyone just thinks of AI as an evil CEO and CFO sit down and say, we're going to cut the agents. Like that's not how the real world works. The real world is struggling to hire enough agents already. So if you give them a button to click to have more agents and if all it took was to pay them 10 or 20 % more and they didn't have to hear about it every day, they probably press that button, right? So the problem is not what people think it is. Like these overcrowded contact centers where people are fearing for their lives about AI automating their livelihoods.

22:33No, like in most places where we implement, people are excited. They start using it. And, you know, those that are excited are the ones that were there for a long time and they tend to stick. The others are just kind of like tourists. They come in for a few months, they leave. And yeah, like, will there be fewer of those in the future? Yep. Will it mean that someone might not have a job they need for a few months? potentially but they'll probably have a job annotating stuff for other companies building on models to do something that we're not even thinking about yet well looking forward to uh watching the the full talk uh in dublin uh at sas stock um i'm curious to know um as a ceo what ai tools you're using maybe like on a day-to-day basis to to uh you know help you do your job make you more efficient make you more productive yeah i mean look i think they're like like ChatGPT has become like my operating system for like absolutely everything.

23:27I was thinking this morning, I used Google for some reason. I was like, oh, I haven't used it in days, right? So I think six months ago, I was talking to people and thinking that I'm probably like 50-50. I literally don't even use Google anymore. I don't think it's a very controversial statement. I think other good tools, I didn't do a lot of coding, but I think Lovable is great. And I think it's great for illustrating kind of like the prototypes and things in product discussions. and then obviously as you know you implement you need a bit more but I think it's really powerful and the team uses it like amazingly cloud code is good granola is a fantastic tool you know I've been a fan of gong for a long time but I think granola is like where and you know this is quite close to what we do so I know the kind of like things they have to do to make things work and I appreciate them right because you know gong has not made some of those things work even though we've been like captive to them and you know gong's quite heavyweight like me joining a one-on-one with you and then like poly-i notaker joining and like it's always been this like pump and unnecessary was with granola and like the ethos of just like notes and stuff i think it's like a really really good tool um and then yeah i think um yeah i i no longer write many lines of code but like cursor is like really heavily used um internally and yeah on chat gbt like what would be some of the use cases you like are you uh using it to draft emails you know uh social posts like memos what would be like a couple of recent use cases that you've used it from a business ceo perspective yeah look i think that outlining long memos of things you know amazon style has never been my forte because i'm a talker i'm not not bad at writing but it requires a context shift that i tend to find very difficult to do.

25:19So it's allowed me to look with, you know, 30 minutes instead of three hours, write out, I don't know, complete new pricing, what, why, how, like 20 pages. I think that's really good. I think in different presentations and stuff like that, it's phenomenal for wordsmithing and all that. Social posts and stuff, it definitely accelerates you. I think there's a definite kind of like degradation in the level of content you see on LinkedIn and stuff. You know, when I write, I'm quite serious about doing a good job. It's very obvious when the tone of the person is not coming through. So in those posts and stuff, I tend to not use it all that much because I think the algorithm will punish you for being yet another, you know, sentences of da-da-da-da-da-da-da.

26:02You know, like it's all sensationalist. It reads like a Daily Mail article. And, you know, it's the number of hyphens, bullet points, and weird emojis is just kind of like whatever. Do you use it for that? You know, I've experimented for sure. So I try and post daily on LinkedIn. I would say most, but not all, if I've used ChatGPT to, like, I'll draft the post, ask it to, like, you know, create a more polished, you know, AI version. and yeah I think you definitely I would say I definitely feel that feel that I'm punished by the LinkedIn algorithm from doing that so it for me when I read it I'm like oh yeah it has improved my post and again I don't know if this is like this like sort of you know psychological like effect there but but in general I would say they don't perform as well but I have seen like people I don't know if you're familiar with like Adam Robinson who runs RB2B and he's like um uh quite uh popular on linkedin uh he would write you know spend a lot of time as a ceo writing like three uh posts uh a week uh but now he's trained uh and he's open about this trained chat gpt to write like him to then buy back all the time that he was spending as a ceo on writing and you know still seems to be performing pretty well but he has a large audience now right so So I think that's what you can lean into.

27:34Listen, I think that it improves. For most people, it vastly improves their content, right? I think actually it's fairly non-controversial for everyone. It can improve it, right? But there is just like this stream of GPT thought where it's like Asians talking to Asians and it's like, okay, you know what? There's a part where you kind of are hungry for like your voice or a grammatical error here and there to just know that someone actually because you know why do we post daily do you have like you know we we know why because we're doing the mangen and all that right but do you genuinely every day have something worthwhile to share with the world i don't know and that's where i think is just you know like the consistency versus um you know i think like the best people at that get excited to the dopamine rush of getting a thousand likes and honestly welcome back to high school right You're waiting for someone to poke you on LinkedIn.

28:32What's next for PolyAI? What's over the next sort of like 6-12 months? And again, in this rapid pace of the AI era, what's next for you guys? Yeah, my North Star is the number of companies for whom we do the work of a thousand people or more. At the moment, we've got three. I'd like that number to be at least double in six months and double again in another six months. and that's really I think the most important growth trajectory because you know who's paying you what and all of that or like how many you captured in the long tail of some kind of like PLG, Zerg Rush, it can be great. It can be a leading indicator for great things or it could just be noise, right?

29:12I think that we're a B2B company and the only real measure of it is like the impact it's had on those companies that decides how sticky they are and really just how good our product is right whether it deserves to be used and to be used by more companies so um you know we're on track for for those things and it's really the goal is just to have a bigger impact on the world um and um yeah that's mostly large enterprises mostly in the us some in the uk a few in continental europe and i think that shape is likely to remain similar for the next kind of like two years well nicola thanks so much for coming on the podcast today uh looking forward to seeing you in Dublin on the 14th of October.

29:56And yeah, really appreciate you sharing with the SaaS.com community your journey or parts of your journey so far and lessons and looking forward to seeing more at SaaS.com. Absolutely. Well, thanks for having me and I look forward to seeing you there. See you there. Thanks for listening to the SaaS Revolution show. If you enjoyed this episode, please leave a review and follow the show. It helps more SaaS and AI founders to discover the podcast and keeps us bringing you the leaders who are shaping the future of the industry. For more insights and to join the SaskTalk community, head to sasstalk.com.